Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
125 Cards in this Set
- Front
- Back
Cross-sectional studies
and diseases with long duration |
Will identify high proportion of prevalent cases with
long duration People with short disease duration may not be identified as diseased If disease duration is associated with exposure, then results can be biased *Example: Those with severe emphysema (outcome) are more likely to smoke (exposure) and have higher case fatality rate (shorter disease duration) than those with less severe emphysema. |
|
Internal Validity
|
Was the study well done?
Are the findings valid? Need to consider… - If there are major methodological problems - If findings could be due to bias, confounding, random error **Important – Need to establish sound internal validity before you consider generalizing the results beyond the study population |
|
External Validity
|
Aka “generalizability” to target population
To what extent are the participants you have studied representative of all people with the outcome of interest? Need to examine… - Who did not participate in the study - Characteristics of study participants that might preclude you from generalizing the study results to others who were not in the study |
|
Clinical Trial
|
Controlled study that prospectively evaluates
the effect of an allocated exposure (i.e., intervention) on the outcome of interest Effects in which we’re interested: Safety, efficacy, effectiveness Considered “gold standard” of epi studies |
|
Examples of Exposures and Outcomes in
Epidemiologic Research |
Exposures:
Medications • Surgical Procedures • Behavior Modification • Screening Programs • Traits and Behaviors • Genetic Variants • Infectious Agents • Environmental Toxins Outcomes • Death • Disease • Subclinical Indicators of Disease • Health-Related Traits • Quality of Life • Physical Function • Costs |
|
Key Parameters of Clinical Trials
|
Individual is unit of observation
Experimental design Follow participants over time -Collect data from at least two time points (e.g., before exposure, after exposure) |
|
When Can a Clinical Trial be Conducted?
|
Clinical trials are justified when uncertainty
exists regarding the effectiveness of a treatment (aka, EQUIPOISE) EQUIPOISE: Legitimate uncertainty or indecision as to choice or course of action… because of an unknown balance of benefits and risks The researcher must believe that… (1) what a study proposes to accomplish has an excellent chance of being helpful (i.e., will contribute to generalized knowledge) and (2) he/she must have justified doubt about the relative benefits of the comparison treatment (which may be the “standard of care” treatment) |
|
When Clinical Trials Are Impossible
(or Nearly Impossible) |
Adverse Exposures (e.g., cigarettes, other
toxins) Rare Outcomes (e.g., Reye’s Syndrome) Intervention Already in Wide Use (e.g., intensive care unit (ICU) medical care) |
|
Basic Protocol in a Clinical Trial
|
1. Obtain approval of Institutional Review
Board (IRB) 2. Enroll participants 3. Gather “baseline” data from participants 4. Allocate exposure to participants 5. Follow-up participants to collect data on outcome 6. Conduct data analyses 7. Report findings |
|
Enrollment of Study Population
|
Enroll a study population with specific
characteristics designed to ensure success of the trial and safety of participants Careful application of inclusion/exclusion criteria Such a study sample will yield the “cleanest” results (high internal validity), but may compromise generalizability (low external validity) |
|
General Inclusion Criteria
|
• Able to provide informed consent
• At high risk for main outcome • At low risk for adverse side effects • No confounding medical conditions • Likely to adhere to treatment and data collection/study procedures |
|
Non-Randomized Allocation of Exposure
|
Clinical Judgment
e.g., Good surgical candidate vs. Not Participant Preference e.g., Wants surgery vs. Doesn’t Alternating First Participant --> Treatment; Next --> Control Day of Week MWF --> treatment; TuThS --> Control ID Numbers Last Digit of SSN Odd --> Treatment; Even --> Control |
|
Why Randomize Exposure Allocation?
|
Ensure that exposure assignment is unbiased
Produce similar groups at baseline by known and unknown factors Goal: any difference between the groups at the end of the study will be the result of the exposure / treatment / intervention Minimizes the threat of selection bias Avoids confounding by indication |
|
Randomized Assignment
|
Unstratified by any variables
– Assignment is completely random – Balanced in the long run, but may be unbalanced in the short run Stratified by key variables –Ensures balance within subgroups defined by key variables before randomization –Stratification variable should be strongly related to outcome (e.g., gender, risk level) |
|
Confounding by Indication in Observational Studies
|
A bias when patients with the worst prognosis
are allocated preferentially to a particular treatment. High risk hypertensive patients are more likely to have adverse outcomes. High risk hypertensive patients are more likely to be prescribed calcium channel blockers (than other drugs hypertensive drugs). Observational studies show that calcium channel blockers are associated with more adverse outcomes |
|
Factorial Design
|
Potentially economical way to test two
treatments simultaneously, if their modes of action are independent OR Method to test for treatment synergy - Is the effect of the combined treatment different than expected based on the effects of the treatments alone? |
|
“Cross Over"
|
Crossing from one treatment group to the other
Unplanned crossover: treatment non-adherence procedures/protocol should be designed to minimize Planned crossover design: administration of treatments one after the other in random (or specified) order treatment may be followed by a “washout” period |
|
Planned “Cross Over”
|
Each participant serves as his/her own control
- creates comparability between treatment groups Feasible only if… - Outcomes are recurrent, and - No “carryover” treatment effect after “washout” period Randomize order of treatments |
|
What is a placebo?
|
A placebo is an inactive or inert intervention or
agent that is given as a substitute for the treatment and where the participant is not informed he/she is receiving the active or inactive intervention |
|
Why Use a Placebo (aka Mask Participant)?
|
Masking subjects to exposure assignment
minimizes information bias Equalizes psychological effects of an “intervention” (aka placebo effect) |
|
The Placebo Effect
|
The effect that is produced by a placebo.
The placebo effect is often measured by comparison of the effect observed in patients receiving the placebo with the effect observed in patients receiving the active treatment. |
|
Possible Differential Placebo Effects
|
Physical qualities of a pill, e.g. red vs. blue, big
vs. small, branded vs. not Device/high tech intervention vs. not Injection vs. oral placebo Headache relief was 6% higher in groups that received subcutaneous (i.e., an injection) placebo versus oral placebo (de Craen et al, J Neurol 2000) |
|
Who to Mask/Blind in the Study and Why
|
Participants: Quantify placebo effects
Physicians: Uniform care apart from study Data Collectors: Uniform outcome ascertainment Data Analysts: Reduce threat of analytic bias |
|
Partial Masking
|
In some circumstances masking of participants
and/or physicians may be impossible or unethical (Surgery, behavior modification) In this setting, others can generally still be masked: Data collectors Adjudicators Laboratory measurements Data analysts |
|
Ascertainment of Outcomes
|
Devise clear, a priori outcome definition(s)
Prefer “harder” to “softer” outcomes Don’t forget about adverse effects/events Mask to the fullest extent possible Standardize methods and equipment Train and certify data collectors Conduct on-going quality assurance |
|
Non-compliance
|
Study Non-compliance:
o Persons who stop participating in study o Do not adhere to protocol Treatment Non-compliance: poop o Persons who do not take all of assigned treatment (i.e., poor adherence) |
|
Approaches to Non-compliance
|
Run-in period / pilot study – randomize subjects
after a trial period assessing compliance Monitor noncompliance: - Interview patients, count pills - Medication bottle devices - Blood or urine tests - Directly observed treatment In the setting of non-compliance, the observed effect will likely be smaller than the true effect |
|
CT Data Analysis Approach
|
1. Intention to Treat (ITT)
2. Treatment received o Observational o No longer have benefits of randomization 3. Subgroup analyses o Small numbers o Hard to determine that treatment effect differs by sub-groups |
|
Intention to Treat (ITT) Approach
|
Analysis by assigned treatment regardless of
the observed course of treatment Maintains initial balance from randomization Highlights problems from adverse effects Conservative approach Strongly recommended as primary approach |
|
Number needed to treat (NNT)
|
Number of patients who would need to be treated
to prevent one outcome NNT = 1 / (outcome frequency in untreated group – outcome frequency in treated group) Small NNT is good Estimates often presented with 95% confidence intervals |
|
Number needed to harm (NNH)
|
Number of patients who would need to be treated
to cause one patient to be harmed (by treatmentrelated adverse events or side effects) NNH = 1 / (adverse event frequency in treated group – adverse event frequency in untreated group) Large NNH is good Estimates often presented with 95% confidence intervals |
|
Safety and Stopping
|
“Stopping rule”
A rule set before the start of the trial that specifies a limit for the observed treatment difference for the primary outcome which, if exceeded, automatically leads to the termination of the treatment or control arm (depending on direction of the difference) |
|
When to stop a clinical trial before
its scheduled end? |
1. Clear evidence of benefit
2. Clear evidence of harm --> Importance of plans to monitor the progress of a trial |
|
Monitoring Progress (Benefits/Harms)
in a Clinical Trial |
Data Safety and Monitoring Board (DSMB):
Independent committee with responsibilities to periodically review accumulated data for evidence of benefit or harm (e.g., adverse events) from the treatment Responsible for making recommendations for modifying the trial, including stopping early, if appropriate |
|
Why CTs Can Be Difficult
|
Hard to find and recruit the right people
Great responsibility on the investigator(s), need for tremendous documentation, cost May take years for outcomes to develop People are free to do as they please: - Some assigned to treatment don’t adhere - Some assigned to control seek treatment - Some drop out of the trial completely (loss-tofollow-up) |
|
Advantages of CTs
|
"Gold standard” (Randomization) of epi studies
Designed to minimize bias “Highest quality evidence available” Results may be combined into systematic reviews |
|
NNT and herd immunity
|
Number needed to vaccinate (NNV)
If vaccine coverage rates are high, vaccination will produce positive herd-immunity effects. Hence, if high coverage rates are attained, the population-based NNV will likely be lower. On the other hand, if coverage rates are low, accounting for herd-immunity effects will have little or no effect on estimates of NNV |
|
Limitations of CTs
|
Cost
Limited external validity - Country, patient characteristics, study procedures, outcome measures Time to conduct and to publish findings Difficult to study rare events Difficult to study distant events Narrowing of the studied question |
|
Phases in Clinical Trials
|
I Evaluate safety, dosage-->10-20 healthy volunteers -->Unexpected side effects may occur
II Evaluate efficacy --> About 200 patients-->Most drugs fail in Phase II due to being less efficacious than anticipated III Evaluate effectiveness--> More than 1,000 patients-->Likelihood to detect rare side effects increases with number of patients IV Evaluate long-term safety and effectiveness--> 1,000s of patients, “real life” evaluation outside of research environment-->Previously untested groups may show adverse reactions, postmarketing surveillance |
|
Definition of Cohort Study
|
Observational epidemiologic study that follows
groups with common characteristics over time Terms associated with cohort studies: followup, incidence, longitudinal study Participants defined by exposure status, then followed for outcomes of interest |
|
Key Parameters of Cohort Studies
|
Individual is unit of observation
Observational design Follow participants over time -Collect data from at least two time points Participants selected based on exposure status, and all are “at risk” for the main outcome at baseline |
|
When is a Cohort Study Warranted?
|
Good evidence of an association of the disease
with a certain exposure Exposure is rare, but incidence of disease among exposed is high Time between exposure and disease is short Attrition of study population can be minimized |
|
Types of Populations in Cohort Studies
|
Open or dynamic:
Changeable characteristic (e.g.,smoking) Members come and go; losses may occur Incidence rate Fixed: Irrevocable event (e.g., birth of child) Does not gain members; losses may occur Incidence Rate Closed: Irrevocable event (e.g.,natural disaster) Does not gain members; no losses occur (observation period is short) Cumulative incidence |
|
Timing of Cohort Studies
|
Prospective – Looking forward in time
Participants grouped based on past or current exposures and followed forward for outcome Retrospective – Looking back in time Both exposures and outcomes have already occurred when study begins, data collection is based on existing records (historical) Ambidirectional – Looking both forward and back in time |
|
Basic Protocol in a Prospective Cohort Study
|
1. Obtain approval of Institutional Review
Board (IRB) 2. Enroll participants grouped by exposure status 3. Gather “baseline” data from participants 4. Follow participants to collect data on outcomes 5. Conduct data analyses 6. Report findings |
|
Selection of Exposed Group (Cohort)
|
Depends on hypothesis, exposure frequency,
feasibility considerations Special cohorts for rare exposures -Uncommon workplace exposures, unusual diets, uncommon lifestyles -Occupational groups, religious groups General cohorts for common exposures -Professional groups or well defined geographic areas Facilitate follow-up and accurate ascertainment of outcomes |
|
Selection of Unexposed Group (Cohort)
|
Internal comparison group
-Unexposed members of same cohort, ideal option General population -Second best option, based on preexisting population data on disease morbidity and mortality Comparison cohort -Members of another cohort, least desirable option, results can be difficult to interpret |
|
Characterizing the Exposure (Cohort)
|
Exposed or index group vs. Unexposed or
referent or comparison group Must specify minimum amount of exposure to qualify as “exposed” May divide exposed group into levels -Example: High, medium, low -Detection of dose response relationship |
|
Things Happen Over Time
Affecting Exposures and Outcomes (Cohort) |
Aging
Environmental exposures -Lifestyle -External environments Disease identification |
|
Cohort Sources of Information
|
Interviews
Medical and employment records Direct physical exams Lab tests and biological specimens Environmental monitoring And remember… Each source has advantages and disadvantages Need comparable procedures for data collection in exposed and unexposed groups, including standard outcome definitions and masking |
|
Follow-up and Outcome Assessment (Cohort)
|
Exposed and unexposed groups monitored
for outcome(s) over time during follow-up More than one outcome is usually studied Follow-up can range from a few hours to a several decades Difficult to maintain contact with participants Aim to have a high “follow-up rate” by minimizing losses to follow-up |
|
Maximizing Participant Retention (Cohort)
|
Losses to follow-up (LTF) decrease sample size
LTF may be more like to develop outcome! Collection of data at baseline on participant, friends, relatives, physicians Regular contact via mail, phone, home visits If possible LTF – Then, “Address Correction Requested,” contacts provided at baseline, directories, national registries, commercial companies |
|
Cohort Study Data Analysis Approach
|
Primary objective – Compare disease
occurrence in exposed and unexposed groups -->Incidence rates, cumulative incidence Person-time Induction period – Interval between action of a cause (e.g., exposure) and disease onset Latent period – Interval between disease onset and clinical diagnosis |
|
Disadvantages of Cohort Studies
|
Inefficient for rare outcomes
Poor info on exposures and other key variables (retrospective) Expensive and time consuming (particularly prospective) Inefficient for diseases with long induction and latent periods (prospective) More vulnerable to bias (retrospective) |
|
Advantages of Cohort Studies
|
Efficient for rare exposures
Good information on exposures (prospective) Can evaluate multiple effects of an exposure Efficient for diseases with long induction and latent periods (retrospective) Less vulnerable to bias (prospective) Can directly measure disease incidence or risk Clear temporal relationship between exposure and outcome (prospective) |
|
Information Bias in Cohort Studies
|
Information bias is a flaw in collecting or
measuring exposure or outcome data that results in different quality/accuracy of information between comparison groups In retrospective cohort studies, you rely on past records What if past records differed in quality and extent of info between exposed and unexposed persons? |
|
Definition of a Case-Control Study
|
Observational epidemiologic study of persons
with the outcome of interest (“cases”) and without (“controls”) that examines the presence of particular attributes (“exposures”) in the two groups Participants defined by outcome status, then exposures of interest are assessed Highly efficient study design |
|
Key Parameters of Case-Control Studies
|
Individual is unit of observation
Observational design No follow-up of participants over time (i.e., the investigator does not directly collect data from the participant over time) - Collect data at one time point Participants selected based on outcome status, then exposures are assessed |
|
Types of Case-Control Studies
|
Population-based:
-Participants identified from within a source population -No pre-existing study infrastructure -Example: Inpatients at Johns Hopkins Hospital today Nested: -Source population is ongoing cohort study -Benefits of cohort and case-control study designs -Example: Participants in ALIVE cohort study |
|
Basic Protocol in a
Case-Control Study |
1. Obtain approval of Institutional Review
Board (IRB) 2. Identify and enroll cases and controls 3. Gather data from participants, records, etc 4. Conduct data analyses 5. Report findings |
|
Selection of Cases (case control) : Case definition
|
Goal – accurate classification
Signs and symptoms Physical and pathological exams Results of diagnostic testing Criteria used have implications for accuracy of case definition Best to use all available evidence Disagreement about how to define disease, disease definitions change over time |
|
Selection of Cases: Sources of Case Identification(case control)
|
Hospital/clinic patient rosters
Death certificates Special surveys (e.g., NHANES) Special reporting systems (e.g., birth and/or death registries) Each source as advantages/disadvantages Goal is to identify as many true cases as possible (and as cheaply and quickly as possible) |
|
Selection of Cases
Incident versus prevalent cases(case control) |
Incidence, if studying causes of disease
Prevalence, if study duration of disease Might not have a choice, so prevalence |
|
Selection of Cases: Complete versus partial case ascertainment(case control)
|
As long as source population can be defined, then…
- Do not have to include all cases in a population -A subset of cases is appropriate |
|
Selection of Controls
Individuals without the disease(case control) |
A sample of the population that produced the
cases AKA, “referent group” because it “refers to” exposure distribution in the source population Cases and controls represent the same base population… So, a member of the control group who gets the disease would end up being a case in the study Selected independent of exposure status! |
|
Assumption (case control)
|
Cases and Controls Originate
From Same Hypothetical Source Population |
|
Selection of Controls(case control)
Sources of Control Identification |
Population (preferred, if available)
Hospital/clinic Deceased individuals Case’s friends, spouse, and relatives Each source has advantages/disadvantages Remember - If a member of the control group actually had the disease, would he/she end up as a case in the study? |
|
Ratio of Controls to Cases
|
Can increase the statistical power of the study
to detect an association by increasing the size of the control group Up to a ratio of 4 controls : 1 case will increase power Beyond 4:1, not considered worthwhile due to costs |
|
Methods of Sampling Controls
in a Nested Case-Control Study |
Each time a case occurs, a control is selected
from the “risk-set” of individuals remaining in the cohort without the condition Cumulative incidence sampling -Risk-set is defined at the end of follow-up among those without the outcome Incidence density sampling -Risk-set is defined at the time of the case -Controls can be future cases! |
|
Challenge in Case-Control Studies
|
Cases and controls may differ in characteristics or
exposures other than the one targeted for study -Is study finding due to exposure, or due to differences between cases and controls? Solution via study design: Match cases and control for factors about which you’re concerned Matching: Process of selecting the controls so they are similar to the cases in certain characteristics (e.g., age, race, sex, etc) |
|
Matching (Case Control)
|
Group matching (frequency matching)
Proportion of controls with a certain characteristic is identical to the proportion of cases with the same characteristic # of controls may be less than # of cases Individual matching (matched pairs) For each case, at least one control is selected who is similar to the case for the characteristic of interest |
|
Problems with Matching (Case Control)
|
Practical –A lot of matching may make it
impossible to find a suitable control Conceptual – Once you match controls to cases by a certain characteristic, then you cannot study that characteristic in your analysis So, only match on factors you are convinced are risk factors for the disease (and you therefore don’t need to investigate) |
|
Sources of Exposure Information(Case Control)
|
Questionnaires: Face-to-face, telephone, self administered/ Can obtain info on may exposures; must be carefully designed and administered to elicit accurate info; expensive
Preexisting records: Administrative, medical, regulatory/ May be only available source of exposure info; avoids bias; may be incomplete; may lack uniformity and details; inexpensive Biomarkers: Levels in blood, urine, bone, toenails/ Estimate of internal dose; infrequently used because of difficultly identifying valid and reliable markers of exposure to noninfectious agents; expensive |
|
Limitations in Recall(Case Control)
|
Virtually all people are limited to varying
degrees in their ability to recall information If this limitation affects all subjects in a study to the same extent, misclassification may result Generally leads to underestimation of the association between exposure and outcome |
|
(Case Control)Recall Bias
|
Differential recall of past exposures/events
between cases and controls -Different by amount of recall -Different by accuracy of recall Can affect study findings on exposure and outcome association |
|
Case-Control Study
Data Analysis Approach |
Challenge: Often, investigators do not know
the size of the total population that produced the cases Assumption: Cases and Controls Originate From Same Hypothetical Source Population So, we don’t know how many people were “at risk” for becoming a case (i.e., we don’t know the denominator), so we can’t calculate incidence, prevalence, associated measures But, we can calculate the odds! Odds of event = probability (p) the event will occur divided by the probability the event will not occur = p / (1-p) |
|
Disadvantages of Case-Control Studies
|
Inefficient for rare exposures
May have poor info on exposures because of retrospective Vulnerable to bias because of retrospective Cannot establish temporal relationship between exposure and disease |
|
Advantages of Case-Control Studies
|
Efficiency
-Less time, less money than cohort studies, experimental studies Efficient for rare diseases Efficient for disease with long induction and latent periods Can evaluate multiple exposures in relation to outcome (so, good for diseases about which little is known) |
|
Measuring the Risk of Disease
|
Absolute risk = Incidence of disease
- No explicit comparison by exposure status Measures of excess risk – Involves comparison by exposure status – Ratio measures (relative risk, prevalence ratio, odds ratio) – Difference measures (incidence difference, prevalence difference) |
|
What is a measure of association?
|
A quantity that expresses the strength or degree
of association (i.e. relationship) between variables. Association: Statistical dependence between two or more events, characteristics, or other variables. In epi studies, the association in which we are interested is between EXPOSURE and OUTCOME IMPORTANT: An association may be fortuitous, spurious, or may be produced by various other circumstances The presence of an association does not necessarily imply a causal relationship |
|
When interpreting a measure of association… We need to evaluate:
|
1. Reference group - The choice of reference group
(‘referent’) 2. Direction - The direction of the estimated measure of association provides information on the nature of the influence of the exposure on the outcome 3. Magnitude - The magnitude of the estimated measure of association provides information about the strength of the relationship between the exposure and outcome |
|
Approaches to Measuring Excess Risk
(i.e., Excess Incidence) in Measures of Assoc |
1. Ratio of Risks:
Risk in Exposed/ Risk in Non-Exposed 2. Differences in Risk: (Risk in Exposed) - (Risk in Non-Exposed) Sound familiar from outbreak investigation calculations? |
|
Choice of the Reference Group
|
When comparing measures of disease frequency,
which group is the referent? Depends on the research question! -The choice may be arbitrary -Often: Choose group with largest sample size -Simple binary exposure: Reference = ‘Unexposed’ When reporting ratio measures of association in tabular format, you typically see the reference group noted by ‘1.0’ or ‘Reference’ or ‘REF’ in a table |
|
Continuous Exposure: What is Reference?
|
Continuous exposure: Measured on interval scale
e.g., Diastolic Blood Pressure, CD4+ cell counts, Age, Smoking Pack-years What is the reference group? -None! -Measures of association reflect increased disease frequency per unit increase in exposure -Need to be clear about units |
|
Case-Control Study
Data Analysis Approach |
Since, we don’t know how many people were
“at risk” for becoming a case (i.e., we don’t know the denominator), we can’t calculate incidence, prevalence, associated measures, like relative risk and prevalence ratio But, we can calculate the odds! Odds of event = probability (p) the event will occur divided by the probability the event will not occur = p / (1-p) |
|
When is the OR a good estimate of the RR?
|
When the cases are representative of all people
with the disease in the population from which the cases were drawn, with regard to history of the exposure. When the controls are representative of all people without the disease in the population from which the cases were drawn, with regard to history of exposure. When the disease is not frequent (i.e., rare). |
|
Attributable Risk For the Total Population
|
If the incidence in the total population is
unknown, it can be calculated if we know: 1. The incidence among exposed. 2. The incidence among unexposed. 3. The proportion of the total population that is exposed. |
|
Reviewing Public Health Surveillance
|
Passive – Routine reporting of disease cases
seen in health care facilities Active - Special search to find disease cases Sentinel – Disease-specific reporting systems in defined catchment areas Syndromic – Uses already existing healthrelated data that precede diagnosis, supplements existing surveillance methods |
|
Basic Concepts of Disease Causation
|
Disease causation is multi-factorial
Pathogenesis is generally a multi-step sequence that may or may not result in clinical disease Agent-host-environment model |
|
Epidemiologic Triangle
|
Agents:
Biologic (e.g., bacteria, virus) Chemical (e.g., poison, smoke) Physical (e.g., injuries) Nutritional (lack and/or excess) Host (factors): Age Sex Genetics Personality (e.g., risk taking) Health status (i.e., previous disease) Immune status Nutritional status Education Environment: (physical) Temperature Crowding Housing Neighborhood Water Sewage Food Radiation Air pollution (Social) Social support Behavioral modeling (e.g., family, peers, culture, media) Politics Economics |
|
“Risk Factor”
|
An attribute associated with the occurrence of
health-related condition - Aspect of personal behavior or lifestyle - Environmental exposure - Inborn or inherited characteristics - Not necessarily causal Several nuanced synonyms: predisposing factor, risk marker, precursor, determinant |
|
Three Essential Characteristics of a Cause
|
Association: a cause must occur together
with its effect; a statistical association must exist between a cause and its putative effect. Time order: a cause must precede its effect. Direction: there is an asymmetrical relationship between the cause and effect; a change in the cause produces a change in the effect; changing the effect does not alter the cause. |
|
Necessary Versus Sufficient Causes
|
Necessary cause = a cause that must
always precede an effect (A Necessary Cause must always precede the effect) Sufficient cause = a cause that always produces an effect Note: The effect need not be the sole result of one cause. |
|
Necessary Causes
|
If a disease is defined by the presence of an
agent, that agent is necessary by definition. Example: Tuberculosis can only be caused by the tubercle bacillus. Contrast: Hepatitis can be caused by many viruses, but Hepatitis C is caused only by the Hepatitis C virus. |
|
Any Given Cause May Be Necessary,
Sufficient, Both, or Neither |
Necessary and sufficient = cause is always
present with disease; nothing but cause is needed to result in disease –Example: measles virus and measles Necessary and not sufficient = cause is always present with disease, but disease is not always present with cause –Example: HPV and cervical cancer Not necessary and sufficient = cause may or may not be present with disease, nothing but cause is needed to result in disease –Example: High-dose exposure to pesticides or ionizing radiation and sterility in men Not necessary and not sufficient = cause may or may not be present with disease; if cause is present with disease, then some additional factor must also be present –Example: sedentary lifestyle and coronary heart disease |
|
Necessary Conditions
|
“X is a necessary condition for Y” =
If we don't have X, then we won't have Y OR Without X, you won't have Y To say that X is a necessary condition for Y does not mean that X guarantees Y. |
|
Sufficient Conditions
|
“X is a sufficient condition for Y” =
if we have X, we know that Y must follow OR X guarantees Y |
|
Epidemiologic Guidelines for Establishing
a Cause-Effect Relationship |
Temporal sequence
Strength of the association Dose response relationship / biologic gradient Consistency of the association / replication Coherence (biologic plausibility) Specificity of the association Experiment (cessation of exposure) Analogy Consideration of alternate explanations |
|
Temporal Sequence
|
Study designs that can establish the potential
“cause” (risk factor or treatment) precedes the disease include: -Clinical trial -Cohort study Study designs that cannot establish that the potential “cause” preceded the disease include: -Cross-sectional study -Population-based case-control study -Ecologic study |
|
Strength of the Association
|
Measures of association include:
-Ratio measures Examples: relative risk, odds ratio Difference measures Examples: risk difference, rate difference, attributable risk The stronger the association between the potential risk factor and the disease, the less likely is the association due to confounding. But, it is possible on occasion for strong associations to be produced by confounding, biases, or chance variability |
|
Probabilistic Causality
|
The strength of a causal relationship is
assessed by the magnitude of its measures of association. The greater the RR or OR, the closer the cause is to being necessary and/or sufficient. |
|
Strength of the Association
Dose-Response Relationship |
Does risk for disease increase
with the degree of exposure? Exception: “threshold effect” |
|
Consistency of the Association
|
“Has it been repeatedly observed by different
persons, in different places, circumstances, and times?” Consistency is persuasive only if the studies use different architectures, methodologies, and subject groups and still come up with the same results. Example: Heart disease and high blood pressure |
|
Coherence of the Association
(Biologic Plausibility) |
“The whole thing should make biologic and
epidemiologic sense.” Caution: This criterion is used to test whether evidence in favor of a hypothesis is plausible based on currently existing theory and knowledge |
|
Specificity of the Association
|
“The precision with which the occurrence of
one variable, to the exclusion of others, will predict the occurrence of another, again to the exclusion of others.” But, the closer one gets to specificity, the easier it is to detect the association of the cause and the disease. The extreme case would be a necessary and sufficient cause for a disease. |
|
Specificity of the Association:
NOT a valid criterion! |
1. “causes” have multiple effects
2. Diseases have multiple causes |
|
When assessing evidence, ask if the
observed association could be… |
1. Due to chance?
2. Due to bias? 3. Due to confounding? |
|
Confidence Interval (CI)
|
A computed interval that, upon repeated
sampling, has a given probability (e.g., 95%) of containing the true value of a statistical parameter (e.g., ratio, proportion, rate). In other words… For a 95% confidence interval, if a single population is repeatedly sampled, then 95% of the samples would capture the true value of the population parameter. Expresses the precision of the point estimate - More narrow interval = more precision - Less narrow interval = less precision Calculated with predetermined significance level, α (alpha), which is often set at 0.05 |
|
Two Principal Types of Bias
|
1. Selection Bias
2. Information Bias |
|
Selection Bias
|
Error due to systematic differences in
characteristics between those who take part in a study and those who do not The problem is that the association between exposure and outcome may differ between those who participate in the study and those who do no The measure of association is distorted due to procedures used to select subjects and from factors that influence study participation Usually inferred, rather than observed |
|
Selection Bias Types
|
Self-selection bias
• Selection of controls – Healthy worker effect • Post-entry exclusion bias |
|
Information Bias
|
A flaw in collecting or measuring exposure
or outcome data that results in different quality/accuracy of information between comparison groups Can result in distortion of the measure of association |
|
Information Bias Types
|
Misclassification
- Differential and non-differential with respect to exposure and outcome status • Recall bias • Reporting bias • Interviewer bias • Surveillance bias / biased follow-up |
|
Bias can distort the measure of association
|
Away from the null (e.g., away from 1.0):
i.e., observe stronger than true association between exposure and outcome True RR = 1.5 Observed RR = 3.2 Toward the null (e.g., toward 1.0): i.e., observe weaker than true association between exposure and outcome True OR = 2.6 Observed OR = 1.2 |
|
Distinguishing between random error (i.e.,
chance) and systematic error (i.e., bias) |
Imagine that a given study could be
increased in size until it was infinitely large Some errors would be reduced to zero; these are the random errors Other errors would not affected by increasing the size of the study; these are systematic errors or bias |
|
Confounding
|
A situation in which the measure of association
is distorted because of the relationship between the exposure and a third factor that also influences the outcome. It is a true phenomenon, and not an error in the study. Distortion in a measure of association due to a third variable that: 1. Is associated with the exposure 2. Influences the outcome 3. Is not in the causal pathway (i.e., not an intermediate step between exposure and outcome) |
|
Consequences of Failing to
Account for Confounding |
In this example, ignoring the confounder (i.e.,
smoking) would have resulted in a “false positive” finding of the association between heavy alcohol consumption and MI. The opposite can also be true, ignoring a confounder can obscure a true association between an exposure and an outcome, leading us to conclude that there is no association when an association truly exists. |
|
Confounding can be controlled for…
|
1. In the study design
• Matching in a case-control study • Randomization in a clinical trial 2. In the data analysis • Stratification • “Adjustment” (e.g., age adjustment) • Multivariate regression models BUT, we must have collected the data! |
|
Interaction (i.e., Effect Modification)
|
If the size of the association between an
exposure and an outcome is changed or modified by the level of a third variable, interaction is said to be present Interaction is also called “effect measure modification” Classic examples: age, immunization |
|
Assessing Effect Modification
|
The presence or absence of an interaction
depends upon the measure of effect that is being used, e.g., risk difference, relative risk Note: If you detect effect modification, then report it via stratum-specific estimates. Do not “adjust” for it in your data analyses. |
|
Model for Additive Effect
|
Combined Total Risk of A and B =
Baseline Risk + Attributable Risk (A) + Attributable Risk (B) Combined Effect of A and B = Attributable Risk (A) + Attributable Risk (B) |
|
Model for Multiplicative Effect
|
)
Combined Total Risk of A and B = Baseline Risk x Relative Risk (A) x Relative Risk (B) Combined Effect of A and B = Relative Risk (A) x Relative Risk (B) |
|
Possible Types of Effect Modification
|
Antagonism: Combined effect less than
predicted by the model (negative interaction) Synergism: Combined effect greater than predicted by the model (positive interaction) |
|
Comparison of
Confounding and Effect Modification |
Confounding – Association between exposure and
outcome is distorted by a third variable related to the exposure and outcome Effect modification – The association between exposure and outcome is modified by levels of a third variable |
|
Distinguishing between
confounding and effect modification |
1. Make list of potential confounders and effect
modifiers (literature review, data collected) 2. Calculate “crude” measure of association for exposure and outcome of interest 3. Stratify association by levels of potential confounder or potential effect modifier 4. Compare crude vs. stratum-specific associations… If stratified associations are relatively similar across strata AND different from crude, then you have confounding If stratified associations differ across strata AND crude association seems to be weighed-average of stratum-specific associations (i.e., crude measure is between stratum-specific measures), then you have effect modification |